Unravelling Bias: A Sardinian perspective on taxonomic, spatial, and temporal biases in vascular plant biodiversity data from GBIF

Biodiversity data are expanding rapidly, yet often exhibit significant biases that are rarely mapped or systematically analyzed to understand underlying drivers. This issue is particularly pressing in the era of citizen science, which now contributes a substantial share of biodiversity records. In this study, we assessed taxonomic, temporal, and spatial biases in vascular plant occurrence records from Sardinia, a Mediterranean biodiversity hotspot, using all available occurrence data for the region retrieved from the Global Biodiversity Information Facility (GBIF). The dataset encompasses a range of sources, from structured inventories to citizen science platforms. Biases were quantified using metrics such as species richness completeness, Pielou's evenness, and the Nearest Neighbor Index (NNI). After mapping these biases, we used Generalized Additive Models (GAMs) to explore their environmental drivers, including road density, the standard deviation of the Normalized Difference Vegetation Index (NDVI), and topographic roughness. Additionally, we evaluated the influence of structured data sources (e.g., Wikiplantbase) versus citizen science platforms (e.g., PlantNet and iNaturalist) on observed bias patterns. Spatial bias was the most prominent, followed by temporal and taxonomic biases. Road density and NDVI influenced both temporal and taxonomic biases, while topographic roughness affected temporal and spatial biases. Structured data mainly contributed to temporal bias, whereas citizen science data were more associated with spatial bias. Our findings highlight the importance of addressing biases in biodiversity data, particularly those introduced by citizen science, and provide a replicable framework for improving data quality and biodiversity monitoring at both sampling and interpretation stage.